Date Published: March 01, 2018
Publisher: International Union of Crystallography
Author(s): Daniel J. Rigden, Jens M. H. Thomas, Felix Simkovic, Adam Simpkin, Martyn D. Winn, Olga Mayans, Ronan M. Keegan.
Novel ways to produce search models from distant homologues for molecular replacement are presented.
Molecular replacement (MR) remains the most popular means to solve the phase problem in macromolecular crystallography. It requires that a search model be placed, usually through sequential rotational and translational searches, in the asymmetric unit in such a way as to provide helpful phase information and allow the calculation of initial electron-density maps (Rossmann & Blow, 1962 ▸). Search models are still obtained predominantly from experimental structures that are recognisably homologous to the target and so are likely to share some degree of structural similarity with it. Some degree of processing of the characterized homologues is often carried out to remove (portions of) side chains or surface loops which sequence comparison shows to be different, or that are likely to adopt a different conformation in the target. Unconventional MR employs different kinds of search models including ideal secondary-structure elements or other regular motifs (Rodríguez et al., 2012 ▸), recurring tertiary packing arrangements (Sammito et al., 2013 ▸), ab initio structure predictions (Bibby et al., 2012 ▸; Keegan et al., 2015 ▸; Simkovic et al., 2016 ▸) and even, for very high resolution cases, single atoms (McCoy et al., 2017 ▸).
It is relatively common to find that a new target bears only distant homology to its closest relative in the PDB. In these circumstances, conventional MR can be time-consuming, dependent on the availability of local expertise, and ultimately unsuccessful, and a wide variety of approaches have been tested to try to automatically extend the range of proteins that can be solved successfully (Keegan & Winn, 2007 ▸, 2008 ▸; Keegan et al., 2011 ▸; Long et al., 2008 ▸; Bunkóczi et al., 2013 ▸; Vagin & Lebedev, 2015 ▸). The work presented here addresses this situation in two ways, based on the recently elucidated correlation between protein structural packing and flexibility and on local rates of evolutionary conservation (Shih et al., 2012 ▸; Yeh et al., 2014 ▸). Firstly, we explore the building of an ensemble of structures based on the distant homologue, using distance geometry methods in CONCOORD. To our knowledge, this is the first such approach whereby a single structure is meaningfully transformed into an ensemble for the purposes of MR, although normal-mode prediction as a means for conformational sampling has a long history of application in MR (McCoy et al., 2013 ▸; Suhre & Sanejouand, 2004a ▸,b ▸). Processing of these CONCOORD-derived structures by AMPLE in the same way that ab initio models are dealt with produces ensemble search models that can solve some exceedingly distant homology cases. Alternative algorithms at each of the two steps involved (ensemble generation and per-residue measurement of structural diversity) could be explored in the future. Secondly, we present a new AMPLE single-structure mode that provides an automated way to sample multiple knowledge-based derivatives of a single distant homologue. The sampling is driven by a per-residue score file, obtained by the user (Supplementary Table S2), which for best performance contains figures reflecting packing or predicted flexibility along the chain. Although performance is less good than with the CONCOORD-derived ensembles, we showed that this can solve targets that were intractable with expertly manually derived structures further edited in various ways by MrBUMP. Several avenues exist for further development. For example, it may be possible to use target–homologue alignments to enable a more sophisticated treatment of side-chain editing. Providing a variable r.m.s.d. error estimate to Phaser that depends on the degree of truncation applied to produce the search model is also worth exploring. Finally, we note that new metrics that may offer improved performance in single-structure editing mode continue to be developed (Liu et al., 2017 ▸).